This Market Isn’t Overvalued, It’s Over-Concentrated

Everyone keeps calling this moment “the new dot-com bubble,” but that comparison misses the real danger. The dot-com crash was chaotic and decentralised: thousands of flimsy startups imploded under their own weight. What we have today is something very different. The giants leading the AI boom, Nvidia, Microsoft, Apple, Google, are not weak, overhyped punts. They’re some of the most profitable firms in history, and that profitability is exactly what makes the system fragile. When Schroders points out that the top five companies now make up roughly 20 percent of the MSCI World Index, double the concentration at the peak of 2000, the implication is simple: almost every investor on earth is leaning on the same handful of shoulders.

The market only works because those shoulders haven’t slipped. Nvidia’s valuation, larger than all UK-listed companies combined, relies on the assumption that demand for AI compute will remain “insatiable,” a term repeated in Morningstar’s bullish semiconductor notes. But insatiable growth is a belief system, not a forecast. It requires a future where nothing unpredictable interferes: no supply chain disruption, no cyberattack, no geopolitical shock. Yet the system behind AI couldn’t be more exposed. The DTCC now ranks geopolitical instability as the number-one financial threat, and S&P Global warns repeatedly that a single fabrication plant outage in Taiwan could stall global AI infrastructure overnight. When a market leans entirely on uninterrupted exponential growth, the acceptable margin for error approaches zero.

What makes this era uniquely fragile is the structure of its financing. The most revealing breakdown comes from Yale Insights, which shows how Nvidia, OpenAI, Microsoft, CoreWeave, and AMD are tied together through a loop of cross-investments that can make demand appear stronger than it is. Capital flows from one firm into another, which then buys GPUs in volumes that feed back into the revenue of the original investor. It’s not fraudulent, it’s strategic, but it blurs the line between organic growth and self-reinforcing momentum. It also creates a single point of failure. If one actor changes behaviour or loses access to capital, the entire loop shrinks instantly.

The historical parallel isn’t the dot-com bubble but the infrastructure providers behind it. Cisco became the world’s most valuable company at the peak of the telecoms buildout, and then lost over 80% of its value when monetisation lagged behind expectation, a dynamic highlighted in Goldman Sachs’ 25-year retrospective. Nvidia has inherited that same position: the essential plumbing of a technology wave whose returns have not yet arrived in proportion to the investment. If monetisation stalls, infrastructure demand reacts immediately. And when the world’s most valuable company is also the most sensitive to a dip in forward expectations, the entire market becomes vulnerable to a single earnings miss.

This fragility is amplified by the broader environment. In 1999, supply chains weren’t geopolitical leverage points, cyber warfare wasn’t a frontline concern, and the world wasn’t split between competing power blocs with conflicting access to semiconductors. Today all of those pressures coexist. When the Economic Times reports that global financial leaders are openly warning about overheating tech markets, it isn’t pessimism, it’s a recognition that a system this concentrated cannot absorb shocks the way the old one could.

The AI boom is not irrational. It’s rational to the point of fragility. Every part of the system is optimised, efficient, interconnected and dependent on the same assumptions about endless compute demand and perfect geopolitical stability. Rational systems don’t wobble gently when stressed. They fail cleanly, suddenly, and without much warning. The danger isn’t that the AI economy resembles 1999. The danger is that it doesn’t. This isn’t a decentralised mania; it’s a centralised architecture with no redundancy. The dot-com crash spread out across thousands of weak points. An AI crash, if one comes, only needs one.

And the world is quietly structured in a way where one is all it takes.